Cross-orientation suppression in human visual cortex

Gijs Joost Brouwer, David J. HeegerJournal of Neurophysiology 2011 · DOI

TL;DR

This study uses fMRI and a novel "forward modeling" approach to investigate how neurons in the human visual cortex (V1) process visual information. They demonstrate that the "normalization model" accurately explains how a grating's response is suppressed by another grating of a different orientation, a phenomenon known as cross-orientation suppression, and that this suppression is implemented as a change in neural contrast gain.

What Problem Does This Paper Address?

Understanding how the brain processes complex visual scenes is a fundamental challenge. Visual cortex neurons don't respond in isolation; their activity is influenced by other neurons and stimuli in their environment. One well-documented phenomenon is "cross-orientation suppression," where a visual stimulus (like a grating) at one orientation can reduce the response of a neuron to a simultaneously presented stimulus at a different, orthogonal orientation. This is psychophysically observed as "cross-orientation masking," where it becomes harder to detect a target stimulus when a distractor stimulus is present.

While a "normalization model" has been successful in explaining these interactions in animal studies (using electrode recordings) and some human fMRI studies, directly testing its predictions in humans, especially for specific neural subpopulations, has been difficult. Standard fMRI measures the average activity of many neurons within a voxel, obscuring the responses of individual orientation-tuned channels. This paper aims to bridge this gap by using a sophisticated fMRI analysis technique to infer the activity of these hypothetical orientation-selective channels in human V1 and test the normalization model's ability to explain cross-orientation suppression.

Methods

The core of this study lies in their innovative use of fMRI combined with a "forward modeling" approach.

  • fMRI Acquisition: Standard fMRI was used to measure brain activity in human participants while they viewed visual stimuli.
  • Weight Estimation Experiment: To establish a baseline of orientation tuning, participants viewed full-contrast gratings at six different orientations. This data was used to estimate "weights" for each voxel, representing the contribution of six hypothetical orientation-selective "channels" to that voxel's overall response. This is akin to mapping the orientation preferences within each voxel.
  • Forward Modeling: This is the key novel technique. Instead of directly analyzing voxel responses, they transformed them into responses of these hypothetical orientation-selective channels. This allowed them to infer the activity of subpopulations of neurons tuned to specific orientations, effectively "decomposing" the voxel signal.
  • Cross-Orientation Suppression Experiment: Participants viewed a vertical target grating at varying contrasts, either alone or superimposed with a horizontal mask grating of fixed contrast.
  • Control Experiment: To isolate the effect of simultaneous presentation, the target and mask gratings were presented sequentially (temporally interleaved) with doubled contrasts.
  • Normalization Model Fitting: The inferred channel responses were then fitted to the normalization model, which predicts how neural responses change with stimulus contrast, incorporating divisive suppression. They compared a version of the model with cross-orientation suppression terms to one without.
  • Psychophysics: Separate psychophysical experiments were conducted to measure cross-orientation masking behaviorally, complementing the fMRI findings.

Datasets

The study utilized data from four healthy human observers. No new benchmark datasets were introduced; the data was collected specifically for this study.

Key Insights and Ideas

  • Inferring Subpopulation Activity: The forward modeling approach is a powerful tool for moving beyond average voxel responses to infer the activity of specific neural subpopulations (defined by orientation tuning in this case) from fMRI data. This opens doors to studying neural interactions in humans with non-invasive methods.
  • Normalization Model Validation: The study provides strong evidence that the normalization model accurately describes cross-orientation suppression in human V1, extending its validity to human fMRI data.
  • Contrast Gain Shift: The suppression is not simply a reduction in response amplitude but a "shift in contrast gain." This means the same amount of stimulus contrast elicits a weaker response when the mask is present, effectively making the neuron less sensitive to contrast.
  • Specificity of Suppression: Cross-orientation suppression is dependent on the simultaneous presentation of stimuli. When the target and mask were presented sequentially, the suppression disappeared, highlighting the importance of temporal overlap.
  • Population vs. Mean Responses: Crucially, the suppression was evident in the inferred channel responses but not in the average V1 voxel responses. This underscores the limitations of traditional fMRI analysis for studying neural interactions and the power of decomposition techniques.

Key Results

  • Channel Responses: For the channel tuned to the target orientation, responses increased with target contrast but were significantly suppressed (rightward shift in contrast-response function) when the orthogonal mask was present.
  • Mask Channel Responses: The channel tuned to the mask orientation showed a baseline response when only the target was present and decreased responses with increasing target contrast when the mask was present.
  • Control Experiment: In the control experiment (temporally interleaved stimuli), no significant shift in contrast gain was observed in the target channel, indicating the absence of suppression.
  • Model Fit: The normalization model with cross-orientation suppression terms provided a significantly better fit to the data from the suppression experiment compared to a model without suppression. Conversely, the suppression-absent model fit the control experiment data better.
  • Psychophysics: Behavioral experiments confirmed that cross-orientation masking occurred only when the target and mask were presented simultaneously, aligning with the fMRI findings.

Limitations and Future Work

  • Voxel Selection: The study selected voxels based on their differential responses to orientations. While this maximized signal-to-noise, it might have excluded voxels with less pronounced orientation tuning.
  • Idealized Channels: The forward model assumes idealized, sinusoidal tuning curves for the hypothetical channels. Real neurons may have more complex tuning profiles.
  • Spatial and Frequency Limitations: The current forward model was limited to orientation and assumed a single spatial frequency channel. Extending it to incorporate multiple spatial frequencies or locations would be necessary for analyzing more complex stimuli.
  • Future Directions: The authors suggest that this approach can be broadly applied to infer responses of various subpopulations of neurons and characterize their interactions across different stimulus dimensions (e.g., color, motion). This opens up exciting avenues for studying how the brain represents complex stimuli and integrates information from different neural populations.

Summary updated

Abstract

Cross-orientation suppression was measured in human primary visual cortex (V1) to test the normalization model. Subjects viewed vertical target gratings (of varying contrasts) with or without a superimposed horizontal mask grating (fixed contrast). We used functional magnetic resonance imaging (fMRI) to measure the activity in each of several hypothetical channels (corresponding to subpopulations of neurons) with different orientation tunings and fit these orientation-selective responses with the normalization model. For the V1 channel maximally tuned to the target orientation, responses increased with target contrast but were suppressed when the horizontal mask was added, evident as a shift in the contrast gain of this channel's responses. For the channel maximally tuned to the mask orientation, a constant baseline response was evoked for all target contrasts when the mask was absent; responses decreased with increasing target contrast when the mask was present. The normalization model provided a good fit to the contrast-response functions with and without the mask. In a control experiment, the target and mask presentations were temporally interleaved, and we found no shift in contrast gain, i.e., no evidence for suppression. We conclude that the normalization model can explain cross-orientation suppression in human visual cortex. The approach adopted here can be applied broadly to infer, simultaneously, the responses of several subpopulations of neurons in the human brain that span particular stimulus or feature spaces, and characterize their interactions. In addition, it allows us to investigate how stimuli are represented by the inferred activity of entire neural populations.
BibTeX
@article{brouwer2011cross,
  author = {Gijs Joost Brouwer and David J. Heeger},
  title = {{Cross-orientation suppression in human visual cortex}},
  journal = {Journal of Neurophysiology},
  year = {2011},
  doi = {10.1152/jn.00540.2011},
}
Figure